[DOsinga]

This reminds of people in the olden days saying that computers can't actually play chess. It's just a well trained excel spreadsheet. But they don't actually play the game.

The article doesn't mention the Turing test. Isn't that the relevant criteria here? Can an expert tell the work in museums from what DALL-E produces?

[ben_w]

It doesn’t take an expert to tell DALL-E 2 from a human artist. I attained a mere GCSE[0] grade C or D in art (can’t remember, it was 22 years ago), and yet I scored 27/30 on this exact question this morning: https://news.ycombinator.com/item?id=31467045

[0] UK exam taken at age 16, which was the age when UK mandatory education ended at the time I took it: https://en.wikipedia.org/wiki/General_Certificate_of_Seconda...

[DOsinga]

Agreed! But that's not really the point the article is trying to make. Once you know DALL-E, it is pretty easy to see which are DALL-E and which are not. Just like you get to know an artist. But in 10 years, if we have 100 DALL-E like networks and you compare them to 100 artists, can we tell the difference then?

[smitty1e]

> saying that computers can't actually play chess. It's just a well trained excel spreadsheet. But they don't actually play the game.

The code has no epistemological grasp of whether it's playing chess or calculating hurricane paths.

We are body, mind, and soul. Externalizing the mental rules of $ACTVITY into silicon for execution does not grant a soul to that silicon.

One might argue that coding chess for a computer is only "good" insofar as it helps the coder understand the problem space, or maybe produces better training tools for humans.

Computation is a means, not an end.